A high-resolution climate change dataset for China is developed. The climatic evaluation results of downscaling suggest as follows. 1) It is feasible to establish one statistical downscaling framework by incorpo-rating the advanced deep learning approach, artificial neural network (ANN), with high robustness. 2) The high-resolution climatology generated by this new method match the observations better than GCMs’ raw outputs. Those large bias induced by GCMs in temperature and precipitation can be reduced from 5°C down to 1°C and from 5 mm down to 0.5 mm, respectively. 3) The future climate projected by this new method have as similar long-term trend as the raw GCM’s results with minor differences in amplitude and spatial pattern. It is estimated that a warmer climate in the whole country with temperature increment at 3?4°C and a wetter (drier) climate in North China (South China) can be expected around 2100 A. D. under RCP8.5 scenario. More scientists can be encouraged to use this new ANN-based downscaling method and this released high resolution climate change dataset can be used to a wide range of communities related to global change studies.

针对中国地区的气候预估问题, 开发一套高时空分辨率(空间分辨率: 0.25°; 时间分辨率: 逐日)的统计降尺度气候变化数据集。降尺度气候预估结果表明: 1) 在传统降尺度方法的基础上引入人工神经网络算法, 开发高时空分辨率的降尺度气候数据, 技术上简便可行; 2) 将这种新方法应用到模式的历史模拟数据上, 温度和降水的气候态偏差显著减小, 其中部分地区的温度偏差可从 5°C减至1°C以下, 降水偏差可从5 mm减至0.5 mm以内; 3) 将这种新方法应用到模式的未来情景数据上, 既能保留气候模式原有的对长期趋势的估计, 又可进一步微调幅度和空间分布, 其中在RCP8.5情景中, 到本世纪末, 中国各地温度普遍升高3~4°C, 降水总量变化不大, 但有北方增多、南方减少的微弱趋势。所提出的基于人工神经网络的统计降尺度方法对温度变量具有一定的普适性, 所开发的高分辨率气候变化数据集可以在其他全球变化相关学科的研究中使用。